JPH07115545A - Image coder - Google Patents

Abstract

PURPOSE:To provide the image coder conducting coding by using a parameter in compliance with an optimum quantization parameter while suppressing increase in the processing time and avoiding a complicated device. CONSTITUTION:The coder is provided with an image input section 10 receiving an image, a coding section 50 coding picture data from the image input section 10, a parameter optimizing section 30 optimizing a parameter for coding processing in the coding section 50, a parameter storage section 40 storing the parameter optimized by the parameter optimizing section 30 and sending the parameter to the coding section 50, and a changeover switch 20 used to select a path for the image data from the picture input section 10 to the parameter optimizing section 30 or the coding section 50 depending whether the parameter is optimised or the coding processing is conducted and attains the parameter optimization independently of the coding processing.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image coding apparatus equipped in a facsimile, an image database or the like, and more particularly to an image coding apparatus for optimizing parameters relating to a coding process.

[0002]

2. Description of the Related Art A technique for improving the coding efficiency and reducing the image quality deterioration by performing an optimization process on each parameter in the coding process according to the characteristics of an image to be actually coded is a known technique. Is.

For example, in Shin Ito, "Basics of Image Information Processing", published by Tokyo University of Science Press, there is a description of optimization regarding transform coding. Specifically, the number of bits used for quantization is distributed according to the variance of each transform coefficient, and an optimum quantization parameter is created. As a result, it is possible to improve the compression rate and image deterioration.

Two conventional examples will be described below.

[Prior Art 1] FIG. 18 is a schematic configuration diagram of a conventional general image coding apparatus. In the figure, 10 is an image input unit, 40 is a parameter holding unit, 50 is an encoding unit, and 60.
Is input image data, 310 is a DCT (discrete cosine transform) conversion unit, 370 is a quantization unit, 380 is a variable length coding unit, 410 is transform coefficient data, 460 is quantized transform coefficient data, and 810 is quantization parameter data. Is.

Next, each part of FIG. 18 will be described. The image input unit 10 inputs an image, and input image data 60
Is sent to the DCT conversion unit 310. DCT converter 310
Performs DCT conversion processing on the input image data 60,
The transform coefficient data 410 is sent to the quantizer 370. The parameter holding unit 40 holds the quantization parameter and sends it as the quantization parameter data 810 to the quantization unit 370 in the encoding unit 50.

The encoder 50 has the following configuration. The quantization unit 370 performs a quantization process on the transform coefficient data 410 according to the quantization parameter data 810, and sends it to the variable length coding unit 380 as quantized transform coefficient data 460. The variable length coding unit 380 uses the quantized transform coefficient data 4
Variable length coding processing is performed on 60.

The operation based on the above configuration will be described. FIG. 19 is a flowchart for explaining the operation of the image coding apparatus in FIG. Based on this, the operation will be described below. First, in S210, an image to be processed is input in the image input unit 10, and is output to the DCT conversion unit 310 as input image data 60. In step S510, the DCT transform unit 310 performs DCT transform processing on the input image data 60, and sends it as transform coefficient data 410 to the quantization unit 370. In S520, the quantization unit 370 quantizes the transform coefficient data 410 according to the quantization parameter 810 presented by the parameter holding unit 40, and sends it to the variable length coding unit 380 as quantized transform coefficient data 460. In S530, the variable length coding unit 380 performs variable length coding on the quantized transform coefficient data 460.

Hereinafter, the image coding apparatus as described above will be referred to as "conventional example 1". Generally, the setting of the quantization parameter in Conventional Example 1 is performed in the following procedure in advance of the encoding process.

Step 1): Select one or more images that are generally assumed to be frequently used.

Step 2): The quantization parameter is optimized for each of the images selected in Step 1).

Step 3): Weighted averaging is performed on each parameter obtained in step 2).

Step 4): The quantization parameter is fixed to the value obtained in Step 3).

In the above-mentioned setting of the quantization parameter, in the procedure 2), the above-mentioned quantization parameter optimization method is used.

The weighted averaging process in step 3) is
It is performed by assuming the appearance probability for each image selected in step 1) and increasing the weight of images with a high appearance rate, or considering the deterioration of image quality and increasing the weight of images of high importance. .

In the conventional example 1 described above, the quantization parameter held in the parameter holding unit 40 is fixed and cannot be changed. Therefore, even when an image having characteristics different from the image assumed in the procedure 1), that is, an image in which the variance of transform coefficients is different, processing is performed with the same quantization parameter. For this reason, depending on the image to be encoded, there are problems that the encoding efficiency is lowered, the image quality is severely deteriorated due to the quantization error and the like. In other words, this means that the effect of the coding method used is not sufficiently obtained.

In order to improve the reduction in the effect of the encoding process, there has been proposed a method of individually designing an optimum quantization parameter by using the above-described optimization method for each image to be actually encoded. .

(Prior Art 2) FIG. 20 is a schematic block diagram of an image coding apparatus for designing optimum quantization parameters for individual images. 18, those parts which are the same as those corresponding parts in FIG. 18 are designated by the same reference numerals, and a description thereof will be omitted. 30 is a parameter optimization unit, 2
10 is a transform coefficient buffer unit, 220 is a control signal, 230
Is transform coefficient data, 320 is an average power measuring unit, 330 is a bit allocating unit, 340 is a parameter creating unit, 420
Is average power data, 430 is bit allocation data, 8
20 is quantization parameter data.

Next, each part of FIG. 20 will be described. The image input unit 10 inputs an image, and input image data 60
Is sent to the DCT conversion unit 310. DCT converter 310
Performs DCT conversion on the input image data 60 and sends it as conversion coefficient data 410 to the average power measuring section 320 in the parameter optimizing section 30 and the conversion coefficient buffer section 210.

The parameter optimizing unit 30 has the following configuration. The average power measurement unit 320 calculates the average power of the transform coefficient data 410 and sends the average power data 420 to the bit allocation unit 330. The bit allocation unit 330 allocates quantized bits based on the average power data 420, and sends it to the parameter generation unit 340 as bit allocation data 430. The parameter creation unit 340 creates a quantization parameter based on the bit allocation data 430 and sends it as the quantization parameter data 820 to the parameter holding unit 40.

The parameter holding unit 40 sends the quantized parameter data 810 to the quantizing unit 370 and sends the control signal 220 to the transform coefficient buffer unit 210. The transform coefficient buffer unit 210 receives the control signal 220,
Transform coefficient data 41 sent from the DCT transform unit 310
Holds 0. When the control signal 220 is received, the transform coefficient data 230 is sent to the quantizer 370 in the encoder 50.
Is sent.

The coding unit 50 has the following configuration. The quantization unit 370 quantizes the transform coefficient data 230 according to the quantization parameter 810, and the quantized transform coefficient data 4
Then, it is sent to the variable length coding unit 380 as 60. The variable length coding unit 380 performs variable length coding on the quantized transform coefficient data 460.

The operation based on the above configuration will be described. 21 and 4 are flowcharts for explaining the operation of the image coding apparatus shown in FIG. Based on this, the operation will be described below.

First, in step S210, an image to be processed is input in the image input section 10 and sent as input image data 60 to the DCT conversion section 310. In step S310, the DCT conversion unit 310 performs DCT conversion processing on the input image data 60, and outputs the conversion coefficient data 410 to the average power measurement unit 320 and the conversion coefficient buffer unit 210. In S320, the transform coefficient buffer unit 210 temporarily stores the transform coefficient data 410. S3
At 30, the average power measuring unit 320 measures the average power of each transform coefficient of the transform coefficient data 410, and sends it as the average power data 420 to the bit allocating unit 330. In step S340, the bit allocation unit 330 allocates bits according to the average power data 420, and the bit allocation data 430 is stored in the parameter generation unit 3.
40 to 40. In S350, the parameter creation unit 34
At 0, the quantization step is determined according to the bit allocation data 430, and the quantization parameter is created. S3
At 60, the quantization parameter data 820 is registered in the parameter holding unit 40. In step S370, the parameter holding unit 40 issues a control signal 22 as an encoding start instruction.
0 is sent to the transform coefficient buffer unit 210. As a result, the transform coefficient buffer unit 210 starts sending the transform coefficient data 230 to the quantization unit 370. In S380,
For the transform coefficient data 230 in the quantizer 370,
Quantization processing according to the quantization parameter data 810 is performed, and the quantization processing coefficient data 460 is transmitted to the variable length coding unit 380. In S390, the variable length coding unit 38
At 0, variable length coding is performed on the quantized transform coefficient data 460.

Hereinafter, the image coding apparatus as described above will be referred to as Conventional Example 2. According to the second conventional example, since the optimum quantization parameter can be used for each image, the maximum effect of the encoding method used for any image can be brought about.

However, the above-mentioned conventional example 2 has two problems as described below.

Problem a): The processing time of the entire encoding process increases by the processing time required for the optimization.

Problem b): A device for optimization processing must be added to the image coding device, which makes the entire device complicated and costly.

The above problem a) is due to the fact that the encoding process cannot be started until then because the quantization parameter is not determined until the optimization process is completed. Problem b)
Is caused by performing the optimization process itself, and specifically refers to addition of the parameter optimizing unit 30, the transform coefficient buffer unit 210, and the like in FIG.

As described above, in the image coding apparatus that does not optimize the quantization parameter as in the first conventional example,
There is a problem that the difference in the effect of the encoding processing such as the compression rate and the image deterioration due to the difference in the distribution of the conversion coefficient of each image is remarkable.

Further, in the image coding apparatus for optimizing the quantization parameter for each image as in the conventional example 2, the processing time is increased and the apparatus configuration becomes complicated as compared with the case where the optimization is not performed. There's a problem.

[0032]

SUMMARY OF THE INVENTION The present invention has been made in view of the above-mentioned circumstances, and uses parameters conforming to optimum quantization parameters while suppressing an increase in processing time and complexity of the apparatus. An object is to provide an image encoding device that performs encoding.

[0033]

An image coding apparatus according to the present invention comprises an image input unit for inputting an image, an encoding unit for encoding image data from the image input unit, and a code for the encoding unit. Parameter optimizing means for optimizing the parameter of the encoding processing, parameter holding means for holding the parameter optimized by the parameter optimizing means and sending it to the encoding means, and performing parameter optimization or encoding processing And a switching means for selectively supplying the image data from the image input means to the parameter optimizing means or the encoding means.

[0034]

The operation of the present invention will be described with reference to FIG.

According to the present invention, the image input unit 10 for inputting an image.
A changeover switch 20 for sending the input image to the average power measuring unit 320 and the quantizing unit 370 in response to switching between the quantization parameter optimizing process and the encoding process; and an average power measuring unit for measuring the average power of the input image. 320, and a bit allocation unit 3 that allocates quantization bits according to the average power
30, a parameter creating unit 340 that creates a quantization parameter according to bit allocation, a parameter buffer unit 350 that temporarily holds the quantization parameter, and an averaging process that performs a weighted averaging process on a plurality of quantization parameters. A processing unit 360, a parameter holding unit 40 that holds the optimum quantization parameter, a quantizing unit 370 that quantizes the input image according to the quantization parameter presented by the parameter holding unit 40, and a variable that performs variable-length coding. The long coding unit 380 is included, and the quantization parameter optimization process is performed before the series of coding processes.

FIG. 1 is a block diagram of the present invention. 20, those parts that are the same as those corresponding parts in FIG. 20 are designated by the same reference numerals, and a description thereof will be omitted. 20 is a changeover switch, 350 is a parameter buffer unit, 360 is an averaging processing unit, 440 and 450 are quantization parameter data.

Next, each part of FIG. 1 will be described. The image input unit 10 inputs an image and sends the input image data 60 to the DCT conversion unit 310. The DCT conversion unit 310 performs DCT conversion on the input image data 60 and sends it as conversion coefficient data 410 to the changeover switch 20. The changeover switch 20 sends the transform coefficient data 410 to the average power measuring unit 320 in the parameter optimizing unit 30 and the quantizing unit 370 in the encoding unit 50 according to the quantization parameter optimization process and the encoding process. To do.

The parameter optimizing unit 30 has the following configuration. The average power measurement unit 320 calculates the average power of the transform coefficient data 410 and sends the average power data 420 to the bit allocation unit 330. The bit allocation unit 330 allocates quantized bits based on the average power data 420, and sends it to the parameter generation unit 340 as bit allocation data 430. The parameter creation unit 340 creates a quantization parameter based on the bit allocation data 430 and sends the quantization parameter data 440 to the parameter buffer unit 350. Parameter buffer unit 3
50 temporarily stores the quantization parameter data 440. The averaging processing unit 360 performs a preset weighted averaging process on the quantization parameter data 450 accumulated in the parameter buffer unit to determine the quantization parameter. Then, the quantization parameter data 82
It is sent to the parameter holding unit 40 as 0.

The parameter holding unit 40 receives and holds the quantization parameter data 820 during the quantization parameter optimization processing. Then, at the time of encoding, the quantization parameter data 810 is sent to the quantization unit 370 in the encoding unit 50.

The coding unit 50 has the following configuration. The quantization unit 370 performs a quantization process on the transform coefficient data 410 according to the quantization parameter data 810, and sends it to the variable length coding unit 380 as quantized transform coefficient data 460. The variable length coding unit 380 uses the quantized transform coefficient data 46.
Variable length coding processing is performed on 0.

2, 3, and 4 are flow charts for explaining the operation of the present invention. The operation will be described below based on these.

FIG. 2 shows an overall rough processing flow in the present invention. The operation in the present invention is divided into two processes, that is, a quantization parameter optimization process and a coding process. In S10, it is determined which of the two processes should be performed. When the quantization parameter is optimized, the changeover switch of FIG. 1 is set to the A side in S20. And S
At 60, the quantization parameter optimization process is executed. S10
When performing the encoding process in step S40, the changeover switch is set to the side B in step S40. Subsequently, an encoding process is performed in S50. The changeover switch is changed by a user or an administrator.

3 (a) and 4 (a) show the operation in the quantization parameter optimization processing S60 of FIG.
FIG. 4B and FIG. 4B are flowcharts for explaining the operation in the encoding process S50 of FIG. In the figure, the work performed by the user or administrator is indicated by dotted lines. The same applies to the following flowcharts. The operation will be described below separately for the quantization parameter optimization process and the encoding process.

First, the operation of the quantization parameter optimizing process will be described with reference to FIGS. 3 (a) and 4 (a). When performing the quantization parameter optimization process, the changeover switch in FIG. 1 is set to the A side.

In S110, the user or the administrator selects one or more images that are estimated to be representative of the images used in this image coding apparatus. Hereinafter, the images selected here will be referred to as a representative image group.

In S120, the image input unit 10 inputs the image selected from the representative image group, and outputs the input image data 60 to the DCT conversion unit 310.

In S130, as the quantization parameter optimization process, the processes of S410 to S440 are performed in a part of the parameter optimization unit 30 and the DCT conversion unit 310. In step S410, the DCT conversion unit 310 performs DCT conversion on the input image data 60 to obtain the conversion coefficient data 410 via the changeover switch 20 set on the A side and to the average power measurement unit 320 in the parameter optimization unit 30. Send out. In S420, the average power measuring unit 320 measures the average power for each conversion coefficient of the conversion coefficient data 410. In S430, the bit allocation unit 330 allocates the number of bits used for data expression to each transform coefficient according to the average power. S44
In 0, the parameter creating unit 340 determines the quantization step based on the bit allocation data, and the parameter buffer unit 3 stores the quantization parameter data 440.
Send to 50.

In S140, the parameter buffer unit 35
At 0, the quantization parameter obtained in S130 is temporarily stored.

In S150, it is determined whether all the images in the representative image group have been processed. If all images have been processed, the process proceeds to S160, and if not, the process proceeds to S120.

In S160, the averaging processing unit 360 performs weighted averaging processing on all the quantization parameters temporarily stored in S140 to create quantization parameters. In the weighted averaging process, preset weighting is performed according to the estimated appearance probability and importance of each image.

At S170, the quantization parameter data 8
20 is sent from the averaging processing unit 360 to the parameter holding unit 40, and the parameter holding unit 40 registers the quantization parameter. This quantization parameter is held until the next quantization parameter optimization process.

Next, based on FIGS. 3 (b) and 4 (b),
The operation of the encoding process will be described. When performing the encoding process, the changeover switch of FIG. 1 is set to the B side.

In S210, the image to be encoded is input in the image input unit 10 and is output to the DCT conversion unit 310 as the input image data 60.

In S220, each unit constituting the encoding unit 50 and the DCT conversion unit 310 are processed from S510 to S510.
Encoding processing is performed in the procedure of 530. In S510, DC
In the T conversion unit 310, D is applied to the input image data 60.
Perform CT conversion. The result is converted coefficient data 41
It is sent to the quantizer 370 in the encoder 50 via the changeover switch 20 as 0. In S520, the quantizer 3
At 70, the transform coefficient data 410 is quantized in accordance with the quantization parameter data 810, and is sent to the variable length coding unit 380 as quantized transform coefficient data 460. In S530, the variable length coding unit 380 performs variable length coding on the quantized transform coefficient data 460.

In the quantization parameter optimizing process in the operations described above, the bit allocation unit 33 is executed in S430.
The bit allocation made by 0 is done as follows. Here, each conversion coefficient is numbered, and the symbol y and the subscript i indicating the number are represented as y _i . Also, the number of conversion coefficients y _i is n, the average power of the conversion coefficients y _i is σ _yi ² ,
Let M be the total number of bits to be assigned. At this time, the number of bits m _i to be allocated to the transform coefficients y _i,

[Equation 1] It is calculated by the formula. Therefore, if only the total number of bits M is set, then the optimum bit allocation can be calculated from the power σ _yi ² of each conversion coefficient. Note that the derivation of the detailed algorithm regarding the optimization processing of the encoding parameter is described in the above-mentioned document “Basics of Image Information Processing”, and thus will be omitted.

As described above, in the bit allocation process, it is necessary to set the total number of bits used for quantization. This total number of bits is determined by the required image quality and coding efficiency.

At this time, the total number of bits may be prepared in several ways for optimization so that the user can appropriately switch the quantization parameter. That is, normally, FIG.
Although only one set of quantization parameters is used as shown in (a), a plurality of sets of quantization parameters prepared in advance may be appropriately switched as shown in FIG. The same effect can be realized by defining the total number of bits by the variable X and creating a variable-containing quantization parameter. FIG. 6C shows such an example. By introducing the variable X in this way, it is possible to substitute the variable into the variable according to the intention of the user at the time of encoding processing, and freely change the parameter for high image quality and high compression rate. You can

By the way, in the above-mentioned quantization parameter optimization process, the selection of the representative image group in S110 is performed such that the average power of the transform coefficient is the average of the transform coefficients of the images frequently used in the specific image coding apparatus. Chosen to be similar to power. For example, in the case of an image encoding device that frequently uses documents, an image mainly composed of characters may be selected as the representative image group.

As described above, by performing the quantization parameter optimizing process according to the type of image that is likely to be used in the image encoding apparatus, highly efficient encoding can be performed.

Next, the quantization parameter optimization process S30
Regarding the above, in the above example, all the images in the representative image group are optimized and then the average is calculated, but an algorithm may be used in which the average is calculated after the optimization of each image is completed. This saves memory for the parameters.

Further, an algorithm for averaging the average powers of the transform coefficients of the respective images, that is, for performing optimization once for the average image, may be used. This can be expressed mathematically as follows. Now, let us say that the representative image group is A, the average power of each image in the representative image group and its transform coefficient is S _A and C _A , and the optimum quantization parameter is P _A, and the average power measurement process for each transform coefficient after DCT transformation is performed. Let g (), f () be the quantization parameter optimization process using equation (1), and <> _A be the weighted averaging process in _A. From this, C _A = g (S _A ) ... (2) In the above explanation of the operation, the algorithm P _A = <f (C _A )> _A (3) for averaging the optimization processing results is presented, but the optimization processing is performed on the average power of the averaged conversion coefficients. The algorithm P _A = f (<C _A > _A ) (4) may be used. When f () is a non-linear function, the results obtained by equations (3) and (4) do not match, but the effect of improving the quantization parameter by f () can be obtained by either equation. Further, strictly speaking, the E = <ρ (f (C A), P A)> The _A ··· (5) comprising parameter optimization error E of the formula, P _A so as to minimize An algorithm of choosing may be used. Here, ρ (x, y) represents the distance between x and y. The distance here may be a distance measured by appropriately weighting each element of both parameters.

FIG. 6 is a comparison between the conventional examples 1 and 2 and the present invention. Along with this, the effects of the present invention will be described below.

According to the present invention, since the quantization parameter optimizing process is performed, the coding process performance such as the coding efficiency and the image quality can be improved as compared with the conventional example 1. This can be explained as follows.

In Conventional Example 1, a group B of the entire image, which is estimated to be general when constructing the apparatus, is assumed, and the quantization parameter is optimized for each image. Then, a weighted averaging process is performed on each obtained parameter to obtain a quantization parameter P
Determine _B. If each image in the image group B is S _B and the average power of its transform coefficient is C _B, and is expressed using the above symbols, the quantization parameter P _B used in this case is C _B = g (S _B ) (6) P _B = <f (C _B )> _B ... (7)

On the other hand, the quantization parameter P _A used in the present invention is obtained by the equation (3). If the image to be actually encoded is x, the optimum quantization parameter P _{x of x is}
Can be expressed as C _X = g (x) ... (8) P _x = f (C _X ) ... (9) Here, considering the definition of the representative image group A and the entire image groups B, Prob {ρ (<C A> A, C X) <ρ (<C B> B, C X)}> Prob {ρ ( <C _A > _A , C _X )> ρ (<C _B > _B , C _X )} (10). Here, Prob {} represents the occurrence probability of the event in {}. Assuming that the function f () is monotonic,
Considering the definitions of P _A and P _B , Prob {ρ (P _A , P _x ) <ρ (P _B , P _x )}> Prob {ρ (P _A , P _x )> ρ (P _B , P _x )} ... (11) This equation (11) generally indicates that P _A is closer to P _x than P _B. Here, it can be said that the assumption that the function f () is monotonic is satisfied from the content of the function f () that assigns the codeword according to the occurrence probability of the pixel value according to the equation (1). Therefore, it can be concluded that the encoding with P _A is superior to that with P _B.

Further, according to the present invention, since the quantization parameter optimizing process and the encoding process are independent, it is not necessary to perform the optimization in parallel with the encoding process as shown in FIG. . Therefore, the processing time for encoding is shorter than that in the second conventional example. At the same time, since the parameter optimizing unit 30 is not necessary during the encoding process, the parameter optimizing unit 30
Can be detachable from the main body 100 of the image encoding device. That is, the wiring between the average power measuring unit 320 and the changeover switch 20 and the wiring between the averaging processing unit 360 and the parameter holding unit 40 are provided with the connection connectors 110 and 120, respectively, and the parameters are optimized during the encoding process. The part 30 can be removed.

As described above, by making the parameter optimizing unit 30 attachable / detachable to / from the main body 100 of the image coding apparatus, the image coding apparatus of the present invention realizes code compression at high speed and with high efficiency. Nevertheless, only a few components such as connectors 110 and 120 for connection are added during the encoding process, and the configuration can be realized with almost the same configuration as the conventional example 1. This simplifies the device,
Cost reduction can be achieved.

The effects of the present invention can be summarized as follows.

(1) As compared with the case where the quantization parameter is not optimized, the image quality is less deteriorated and the coding with a high compression rate can be performed.

(2) Despite the optimization of the quantization parameter, the processing time required for the encoding does not become longer than that of the image encoding apparatus which does not optimize the quantization parameter.

(3) At the time of encoding processing, it can be realized with the same simple configuration as that of an image encoding apparatus that does not optimize quantization parameters.

[0072]

EXAMPLES In the following, in the present invention, [1] a case of applying to adaptive coding [2] a case of simplifying selection of a representative image group by automation [3] a case of generalizing the present invention [4] [ Generalized Example 1) [5] Five examples of generalized example [2] will be described.

[First Embodiment] Here, a case where the present invention is applied to adaptive coding will be described as a first embodiment. The adaptive coding here means that the image is divided into small areas, and each area is classified into a category according to its image characteristics.
This is a well-known technique for performing encoding using a quantization parameter suitable for each category. Examples of categories include those that use two categories, the character part and the image part.
For example, the technique described in Japanese Patent Laid-Open No. 4-87460 corresponds to this. Since the DCT performs quantization in the frequency space, using unique quantization parameters for different categories of frequency characteristics greatly contributes to the improvement of coding efficiency.

Hereinafter, the word category is used in this sense. Further, when there are a plurality of categories, a plurality of quantization parameters are required, and this quantization parameter group will be called a parameter set.

FIG. 7 is a block diagram showing the first embodiment of the present invention. In the figure, the same parts as those in FIG. Reference numeral 510 is a region dividing unit, 520 is a category discriminating unit, 610 and 630 are region image data, and 620.
Is category discrimination data, 640 is region conversion coefficient data,
Reference numeral 650 is area quantized transform coefficient data.

Next, each part of FIG. 7 will be described. The image input unit 10 inputs an image and sends the input image data 60 to the area dividing unit 510. The area division unit 510 divides the input image data 60 into small areas by a preset method, and sends the area image data 610 and 630 to the category determination unit 520 and the DCT conversion unit 310, respectively. The category discriminating unit 520 classifies the area image data 610 into any of predetermined categories. Then, the classified result is sent to the changeover switch 20 as the category discrimination data 620. DCT converter 310
Performs DCT conversion on the area image data 630 and sends it to the changeover switch 20 as area conversion coefficient data 640.

The changeover switch 20 receives the domain transform coefficient data 640 from the average power measuring section 320 in the parameter optimizing section 30 and the quantizing section in the encoding section 50 according to the quantization parameter optimization processing and the encoding processing. To 370. At the same time, the category discrimination data 620 is also sent to the parameter optimizing unit 30 and the parameter holding unit 40 in accordance with the switching between the optimizing process and the encoding process. The changeover switch 20 includes two switches, both of which are interlocked. Therefore, it does not happen that one is connected to A and the other is connected to B.

The parameter optimizing unit 30 has the following configuration. The average power measuring unit 320 uses the area conversion coefficient data 62
The average power of 0 is calculated, and the average power data 420 is sent to the bit allocation unit 330. Bit allocation unit 330
Assigns quantized bits based on the average power data 420 and sends it to the parameter creating unit 340 as bit assignment data 430. Parameter creation unit 340
Creates a quantization parameter based on the bit allocation data 430 and sends the quantization parameter data 440 to the parameter buffer unit 350. The parameter buffer unit 350 temporarily stores the quantization parameter data 440 and sends it as the quantization parameter data 450 to the averaging processing unit 360. The averaging processing unit 360 performs weighted averaging processing for each category on the quantization parameter data 450. Then, the quantization parameter data 820
Is sent to the parameter holding unit 40.

The parameter holding unit 40 receives the quantization parameter data 820 for each category during the quantization parameter optimization processing, and holds all of them as a parameter set. A parameter corresponding to the result of the category discrimination data 620 is selected from the parameter set held at the time of encoding processing, and is transmitted to the encoding unit 50 as the quantization parameter data 810.

The encoder 50 has the following configuration. The quantization unit 370 performs a quantization process on the area transform coefficient data 640 according to the quantization parameter data 810, and sends it to the variable length coding unit 380 as area quantized transform coefficient data 650. The variable length coding unit 380 performs variable length coding on the area quantized transform coefficient data 650.

2, FIG. 4A, FIG. 4B, FIG. 8 and FIG. 9 are flow charts for explaining the operation of this embodiment. The operation will be described below based on these.

The operation shown in FIG. 2 is the same as that described in FIG.

FIGS. 4 (a) and 8 and FIGS. 4 (b) and 9 show
The quantization parameter optimization processing S60 of FIG.
It is a flow chart explaining operation in coding processing S50. Hereinafter, the operation will be described separately for the quantization parameter optimization process and the encoding process.

First, the operation of the quantization parameter optimization process will be described with reference to FIGS. S1
In 10, the representative image group is selected by the same procedure as described in the section of action. In S120, the image input unit 10 inputs the image selected from the representative image group, and sends the input image data 60 to the area dividing unit 510. S605
Then, the area dividing unit 510 divides the input image data 60 into areas by a preset method. S610
Then, the area dividing unit 510 selects one area that has not been used for the optimization processing, and the area image data 610, 6 is selected.
The data is sent to the category discriminating unit 520 and the DCT transforming unit 310 as 30, respectively. In S620, the category discrimination unit 52
At 0, category discrimination processing is performed on the area image data 610. Then, the discrimination result is the category discrimination data 6
It is sent to the average power measuring unit 320 in the parameter optimizing unit 30 via the changeover switch 20.

In S130, as the quantization parameter optimization process, the parameter optimization unit 30 and the DCT conversion unit 310 perform the processes of S410 to S440 in FIG. 4A. Descriptions of these are the same as those in the operation section, and thus will be omitted. However, as a difference from the explanation of the section of action, it is assumed that each data passed therein always includes category discrimination data.

In S630, it is determined whether or not the entire area of the input image data 60 has been processed. If all areas have been processed, proceed to S140, otherwise S61.
Return to 0. In S140, the parameter buffer unit 350
At, the quantization parameter data 440 obtained in S130 is divided into categories and temporarily stored. S15
At 0, it is determined whether all the images in the representative image group have been processed. If all images have been processed, the process proceeds to S640, and if not, the process returns to S120. In S640, the averaging processing unit 360 performs weighted averaging processing for each category for all the quantization parameters temporarily stored in S140. Then, the created quantization parameter is sent to the parameter holding unit 40 as the quantization parameter data 820. In S170, the quantization parameter is registered in the parameter holding unit 40 as described in the section of the operation.

Next, the operation of the encoding process will be described with reference to FIGS. 4B and 9. In S210, the image input unit 10 inputs an image to be encoded, and outputs the image as input image data 60 to the area dividing unit 510. S6
In 05, the area dividing unit 510 inputs the input image data 6
Divide 0 into regions by a preset method. S6
10, the area dividing unit 510 selects one area that has not yet been used in the optimization processing, and the area image data 61
0 and 630 respectively, the category discrimination unit 520 and DC
It is sent to the T conversion unit 310. In step S620, the category determination unit 520 performs category determination processing on the area image data 610, and the determination result is the category determination data 6
It is sent to the parameter holding unit 40 as 20.

In S220, as the encoding process, in the encoding unit 50 and the DCT conversion unit, S5 of FIG.
The processing from 10 to S530 is performed. Descriptions of these are the same as those in the operation section, and thus will be omitted. However, as a difference from the explanation of the action section, the quantization parameter data 81
It is assumed that 0 is determined by the parameter holding unit 40 as a parameter for the category indicated by the category discrimination data 620.

In S630, it is determined whether or not the entire area of the input image data 60 has been processed. If all areas have been processed, the process ends, otherwise S610.
Return to.

As described above, the flow of processing at the time of encoding processing is exactly the same as that of ordinary general adaptive encoding. Therefore, there is no increase in processing time at the time of encoding processing by the quantization parameter optimization processing according to the present invention.

In the above operation, the category discriminating data 620 is required when the optimum quantization parameter is created in S130 in the quantization parameter optimization process, because the quantization parameter is optimized for each category. This category-wise optimization makes it possible to create optimum quantization parameters for each category.

At this time, the average power data 420, the bit allocation data 430, the quantization parameter data 44
It is necessary to add to 0, 450, and 820 information indicating which category is the parameter obtained for the image included. Therefore, a method such as adding header information is used. For example, a data format as shown in FIG. By adding category information to the quantization parameter 440, the storage unit of the parameter buffer unit 350 can be organized as shown in FIG.
Further, if category information is added to the quantization parameter 450, weighted averaging processing for each category can be realized in the averaging processing unit 360. Then, by adding category information to the quantization parameter 820, the parameter holding unit 40 can hold the optimum quantization parameter for each category as shown in FIG.

As a method of realizing the processing for each category without using these category data, the category discrimination data 620 may be directly input to the averaging processing section 360. In such a structure, it is necessary to pay attention to the timing of category data transfer.

The category discrimination method in S620 can be realized by a known technique as described in, for example, the above-mentioned Japanese Patent Laid-Open No. 4-87460.

According to the first embodiment, the following effects can be obtained.

(1) Compared with the case where the quantization parameter is not optimized in the adaptive coding process, the compression rate can be increased and the image quality is less deteriorated with the same code amount.

(2) The processing time required for encoding is the same as the processing time for encoding in which the quantization parameter is not optimized in the adaptive encoding process.

(3) At the time of encoding processing, it can be realized with the same configuration as that of an image encoding apparatus that does not optimize quantization parameters.

Second Embodiment Here, as a second embodiment, a case where the selection of the representative image group in the quantization parameter optimization processing of the present invention is simplified will be described.

In the method described in the operation of the present invention and the first embodiment, the selection of the representative image group is performed by the user or the administrator. In this embodiment, automation is attempted to reduce the load of this work. The standard for selecting the representative image group is generally the frequency of use and the degree of importance. In other words, there are two criteria: standard a) selecting a representative image group from frequently used images and standard b) selecting a representative image group from images of high importance. In this embodiment, automation is performed for the case of reference a).

In order to realize the criterion a), the statistics of the manuscript used in a specific image coding apparatus may be taken. In other words, during the quantization parameter optimization process, the originals used by the image coding apparatus up to that point are regarded as the representative image group. Details of this execution method will be described later. According to this policy, the following algorithm is used.

1): During the encoding process, the quantization parameter is optimized independently of the encoding 2): As a result, the obtained quantization parameter is saved 3): 1), 2) 4): Obtain the average of 2) as the optimum quantization parameter. The image group estimated to be the representative image group by the above process may not be appropriate because the criterion b) is not considered. In order to correct this, it is necessary to input the image group of the difference between the representative image group estimated by the above processing and the desired representative image group to the image coding apparatus. Hereinafter, the representative image group based on the estimation is referred to as an estimated representative image group, and the correction image group is referred to as an additional image group.

FIG. 11 is a block diagram showing the second embodiment of the present invention. In the figure, the same parts as those in FIG.

Next, each part of FIG. 11 will be described. The image input unit 10 inputs an image, and input image data 60
Is sent to the DCT conversion unit 310. DCT converter 310
Performs DCT conversion processing on the input image data and sends it as conversion coefficient data 410 to the changeover switch 20 and the parameter optimizing unit 30. The changeover switch 20 sends the input image data 60 to the quantization unit 370 in the encoding unit 50 only during the encoding process.

The parameter optimizing unit 30, the parameter holding unit 40, and the encoding unit 50 are the same as those in the means for solving the problems, and therefore will be omitted.

2, FIG. 12 and FIG. 13 are flow charts for explaining the operation of this embodiment. The operation will be described below based on these.

The operation shown in FIG. 2 is the same as that explained in FIG.

12 and 13 are flowcharts for explaining the operations in the quantization parameter optimizing process S60 and the encoding process S50 of FIG. 2, respectively. Hereinafter, the operation will be described separately for the quantization parameter optimization process and the encoding process.

First, the operation of the encoding process will be described with reference to FIG. In S210, the image to be encoded is input in the image input unit 10, and is output to the DCT conversion unit 310 as the input image data 60. The following S220 and S130 are processed in parallel.

At S220, the encoder 50 and the DCT
In the conversion unit 310, steps S410 to S410 shown in FIG.
The processing of 440 is performed. Further, in S130, in the parameter optimization unit 30 and the DCT conversion unit 310,
The processing from S510 to S530 shown in FIG. 4B is performed. Since these explanations are the same as those in the operation section, they are omitted. Note that both S410 and S510 are DCT conversion processing, but in the present embodiment, this processing is common and may be executed only once.

In S140, the quantization parameter data 440 created in S130 is stored in the parameter buffer unit 3
Temporarily store at 50.

Next, the operation of the quantization parameter optimization process will be described with reference to FIG. In S710, the necessity of the additional image group is determined. If not necessary, the process proceeds to S160, and if necessary, the process proceeds to S720. In S720, an additional image group is selected. For example, the estimated representative image group is corrected by selecting an image that is desired to avoid deterioration as much as possible. In S730, one of the images in the additional image group is selected and input in the image input unit 10.

In S130, in the parameter optimizing unit 30 and the DCT transforming unit 310, S shown in FIG.
The processing from 510 to S530 is performed. This explanation is the same as that of the operation section, and thus will be omitted.

In S140, the quantized parameter data 440 created in S130 is stored in the parameter buffer unit 3
Temporarily store at 50. In S150, it is determined whether or not there is an unprocessed image in the additional image group. If there is an unprocessed image, the process proceeds to S730, and if not, the process proceeds to S160. In S160, the averaging processing unit 360 performs averaging processing on the quantized parameter data 450 accumulated in the parameter buffer unit 350. Normal averaging is performed on the quantization parameter obtained at the time of encoding processing, and weighted averaging is performed on the additional image group if specified. In S170, the parameter holding unit 40 registers the quantization parameter data 820. This data is retained until the next optimization process is performed.

In the above operation, the encoding process for the input image and the quantization parameter optimizing process are performed independently, so that the time required for the encoding process does not increase.

The validity of the estimated representative image group used above is explained as follows.

The representative image group to be created according to the above-mentioned criterion a) is ideally a group A _x of all the images coded by the image coding apparatus from this time to the next optimization processing. Since it is theoretically impossible to obtain A _x ,
This A _x is approximated by the estimated representative image group. Therefore, the document collection A _y used in the image coding apparatus from the previous optimization processing to the present is used as an approximation of A _x . That is, this A _y becomes the estimated representative image group. When g () is the function for obtaining the average characteristic of the image from the image group, and g (A _x ) = g (A _y ) ... (12), this approximation can be said to be valid. In other words, when comparing the previous optimization process to the present and the present to the next optimization process, if there is not much variation in the average characteristics of the image group used in this image encoding device,
This approximation works well.

At this time, the interval for performing the optimization process is free. Of course, they need not be evenly spaced or continuous. If this interval is short, the quantization parameter changes sensitively, and if the interval is long, the change becomes stable. In addition, A _y may include all previously encoded images. Alternatively, when the estimated representative image group used last time is A _y1 and the estimated representative image group obtained this time is A _y2 , A _y = α · A _y1 + β · A _y2 (α, β> 0) (13) A _y may be determined as follows. This stabilizes the change in the quantization parameter. The above conditions are preferably set so that A _y is a good approximation of A _x .

The parameter optimizing unit 30 may be removable as in the first embodiment. However, in this case, the characteristics of the image to be encoded cannot be recorded while the parameter optimizing unit 30 is removed.

According to the second embodiment, the following effects can be obtained.

(1) As compared with the case where the quantization parameter is not optimized, the image quality is less deteriorated and the coding with a high compression rate can be performed.

(2) The processing time required for encoding is the same as that of an image encoding apparatus that does not optimize quantization parameters.

(3) The load on the process of selecting the representative image group in the present invention can be reduced by automation.

(4) When the encoding process is performed, if the optimization process for creating the estimated representative image group is not performed, it can be realized with the same configuration as the image encoding device that does not perform the optimization.

[Third Embodiment] The present invention can be applied to other than transform coding using DCT. For example,
Nobuhiko Fukibuki, “Image signal processing for fax and office automation”,
The publication of Nikkan Kogyo Shimbun Co., Ltd. describes allocation of code words in entropy coding such as Huffman coding. According to this, the allocation of codewords affects the compression rate, but the optimal allocation depends on the characteristics of the object to be coded. Specifically, when a codeword having a length log ₂ (1 / P) is given to an event having an occurrence probability P, the average length of the entire codeword becomes the shortest. For an image, the codeword length is determined according to the probability of occurrence of pixel values.

As described above, in the encoding process, there are parameters that affect the encoding performance. Among these, those that can be optimized according to the characteristics of the image will be referred to as coding parameters. The example of the DCT coding described above corresponds to the optimization of the coding parameter called the quantization parameter with respect to the image characteristic called the average power of the transform coefficient. Further, in the case of the codeword allocation of the entropy coding described above, the occurrence probability of each pixel value is the characteristic of the image, and the codeword length corresponds to the coding parameter.

Of these encoding parameters, the present invention can be applied when the relationship between the image characteristics and the optimization processing is a monotonous function. Function f mentioned in the action section
If () is not monotonic, the above argument does not necessarily hold.

FIG. 14 is a general configuration diagram of the present invention. In the figure, the same parts as those in FIG. 70 and 80 are encoded parameter data. Further, FIG. 15 is a flowchart for explaining the operation when the present invention is generalized. Details of each are similar to the description of the present invention in FIG. The generalization of the present invention in this way has the following effects.

(1) Coding with higher coding efficiency can be performed as compared with an image coding apparatus using fixed coding parameters.

(2) When the lossy encoding method is used,
As compared with an image encoding device that uses fixed encoding parameters, good image quality encoding can be performed with the same code amount.

(3) The processing time at the time of encoding is not increased by the optimization processing of the encoding parameters.

(4) The structure for the coding parameter optimization process is not required at the time of the coding process, and the structure can be simplified.

[Fourth Embodiment] Consider generalization of the first embodiment after the third embodiment. FIG. 16 is a schematic configuration diagram thereof. Since the operation and effect are similar to those of the first embodiment, they are omitted.

[Fifth Embodiment] Consider generalization of the second embodiment following the third embodiment. FIG. 17 is a schematic configuration diagram thereof. Since the operation and effect are similar to those of the first embodiment, they are omitted.

[0135]

As is apparent from the above description, according to the present invention, the image coding apparatus optimizes the quantization parameter in the DCT transform coding, so that it is more efficient than the conventional apparatus that does not perform the optimization. It is possible to perform high compression rate / high image quality encoding processing. Further, compared with the conventional image coding apparatus that performs the quantization parameter optimization processing, the processing load, the apparatus configuration, and the like can reduce the load during the coding processing. Furthermore, the same effect can be obtained by applying the present invention to another encoding method.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing a concept of the present invention.

FIG. 2 is a flowchart for explaining an overall rough flow of processing in the image encoding device of the present invention.

3 is a flowchart illustrating a quantization parameter optimizing process and an encoding process in the flowchart of FIG. 2, respectively.

4 is a flowchart for explaining optimization and encoding of quantization parameters in the flowchart of FIG. 3, respectively.

FIG. 5 is an explanatory diagram showing an application example of quantization parameter expression in the present invention.

FIG. 6 is an explanatory diagram showing a comparison between the present invention and a conventional example.

FIG. 7 is a configuration diagram showing a first embodiment of the present invention.

FIG. 8 is a flowchart for explaining the operation in the first embodiment.

FIG. 9 is a flowchart for explaining the operation in the first embodiment.

FIG. 10 is an explanatory diagram showing an example of a data format of an encoding parameter in the first embodiment.

FIG. 11 is a configuration diagram showing a second embodiment of the present invention.

FIG. 12 is a flowchart for explaining the operation of a quantization parameter optimization process in the second embodiment.

FIG. 13 is a flowchart for explaining operations of quantization parameter optimization and encoding in the second embodiment.

FIG. 14 is a configuration diagram showing a third embodiment of the present invention.

FIG. 15 is a flowchart for explaining an operation in the third embodiment.

FIG. 16 is a configuration diagram showing a fourth embodiment of the present invention.

FIG. 17 is a configuration diagram showing a fifth embodiment of the present invention.

FIG. 18 is a schematic configuration diagram of a first conventional example.

FIG. 19 is a flowchart illustrating an operation in the first conventional example.

FIG. 20 is a schematic configuration diagram of a second conventional example.

FIG. 21 is a flowchart illustrating an operation in the second conventional example.

[Explanation of symbols]

10 ... Image input unit, 20 ... Changeover switch, 30 ... Parameter optimization unit, 40 ... Parameter holding unit, 50 ... Encoding unit, 60 ... Input image data, 70, 80 ... Encoding parameter data, 100 ... Image encoding Device body, 11
0, 120 ... Connection connector, 210 ... Transform coefficient buffer unit, 220 ... Control signal, 230 ... Transform coefficient data, 31
0 ... DCT conversion unit, 320 ... Average power measurement unit, 330 ...
Bit allocation unit, 340 ... Parameter creation unit, 350
... parameter buffer unit, 360 ... averaging processing unit, 37
0 ... Quantization unit, 380 ... Variable length coding unit, 410 ... Transform coefficient data, 420 ... Average power data, 430 ... Bit allocation data, 440, 450, 810, 820 ... Quantization parameter data, 460 ... Quantization transform Coefficient data, 510 ... Area dividing unit, 520 ... Category determining unit, 6
10 ... Area image data, 620 ... Category discrimination data,
630 ... region image data, 640 ... region transform coefficient data, 650 ... region quantized transform coefficient data

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Ken Umezawa 3-7-1 Fuchu, Iwatsuki City, Saitama Prefecture Fuji Xerox Co., Ltd.

Claims (6)

[Claims] 1. An image input unit for inputting an image, an encoding unit for encoding image data from the image input unit, and a parameter optimizing unit for optimizing parameters of an encoding process in the encoding unit. , Parameter holding means for holding the parameter optimized by the parameter optimizing means and sending it to the encoding means, and image data from the image input means depending on whether parameter optimization or encoding processing is performed. An image coding apparatus comprising: the parameter optimizing means or a switching means for selectively supplying to the coding means. 2. The encoding means is capable of switching a parameter used for encoding in accordance with an image to be encoded and an application, and the parameter optimizing means is one of parameters to be switched by the encoding means. The image coding apparatus according to claim 1, wherein optimization is performed for all of them. 3. The encoding means expresses a parameter used for encoding by a variable, and the parameter optimizing means optimizes a parameter including the variable. The image coding apparatus according to claim 1. 4. The encoding means is capable of adaptively switching parameters used for encoding in the same image, and the parameter optimizing means is adaptive in the same image by the encoding means. The image coding apparatus according to claim 1, wherein the optimization is performed for all of the parameters switched to. 5. The parameter optimizing means approximates a representative image used in the optimizing process with an image group coded up to that point in time by the image coding apparatus. The image coding device according to item 1. 6. The image coding apparatus according to claim 1, wherein the parameter optimizing unit is attachable to and detachable from a main body of the image coding apparatus.